<p>Recent evidence within the noisy-channel framework suggests that when encountering implausible sentences, rational readers can make inferences about the speaker’s intentions. However, it remains unclear how the nature of noise and the degree of changes from a plausible sentence to its implausible alternative affect the reader’s inferences. Using an explicit error-correction task, we manipulated the degree of changes in implausible sentences by transposing two characters within a Chinese reversible word, either maintaining or changing its syntactic category. The nature of noise was also manipulated between participants by having them read, along with experimental stimuli, exposure sentences that contained common mistakes, transposition errors, mixed errors, or no errors. Results showed that Chinese readers were able to infer that the implausible test sentences were corrupted by word transpositions, and that their inferences were modulated by the exposure condition, but not by the degree of changes. These findings are consistent with the Context-Specific Noise hypothesis, suggesting that Chinese readers make fine-grained inferences for implausible sentences, independent of the degree of changes between the implausible and plausible alternatives.</p>